For success in the structural genomics initiatives, it is essential that the methods be as powerful, as reliable, and as automated as possible. In work on our new program, Phaser, we have been applying likelihood-based methods to solving the phase problem of X-ray crystallography by both molecular replacement and experimental phasing. Likelihood-based methods have two advantages in the context of automation. First, they are more powerful and therefore have a greater convergence radius. Second, likelihood is a natural scoring function to measure success in satisfying the data, which makes it an excellent criterion for automated decision-making. We will enhance the automation features that already exist in the molecular replacement and experimental phasing modules of Phaser so that, as much as possible, only the diffraction data and sequence information are needed as input. We will work to increase the sophistication and power of our methods, by taking account of more of the correlations that have been neglected to date in order to reduce the computational demands. We will also work to make our algorithms easily accessible to other Projects within PHENIX, by using the Boost.Python library to make them available from the Python scripting language that underpins PHENIX. This will allow the development of automated pipelines combining algorithms from all the components of PHENIX. With the completion of the human genome, there is a new focus on finding ways to exploit the wealth of information it has revealed. In many cases, knowing the 3D structure of the proteins encoded by the genes is invaluable in understanding how they work together, and the primary method to determine those structures is the method of X-ray crystallography. Our work on applying statistical methods to 3D structure determination is helping to make those methods faster, more automatic and more powerful.